Atlas-based analysis methods rely on the morphological similarity between the atlas and target images, and on the availability of labelled images. Problems can arise when the deformations introduced by pathologies affect the similarity between the atlas and a patient's image. The aim of this work is to exploit the morphological dissimilarities between atlas databases and pathological images to diagnose the underlying clinical condition, while avoiding the dependence on labelled images. To this end, the proposed method is formulated under the principle of atlas-based segmentation but, instead of propagating labels, we propagate image intensities. Using a set of MR atlas databases, each database associated with a clinical condition, synthetic images are generated for a target image. The diagnosis is established by assessing the degree of similarity between the synthetic images and the target and assigning the condition of the top-ranked synthetic image. The obtained results are comparable to state-of-the-art methods using annotated images, with an accuracy of 90.0% when evaluated on a set of 60 whole heart MR images containing healthy and pathological subjects.